Menu
AboutModelsLeadershipDysthroverseResearchSafetyJoin

MODEL ANNOUNCEMENT / OCTOBER 8, 2026

Introducing
Whisper 7.3

Small model. Premature conclusions.

Introducing Whisper 7.3, our fastest route from an incomplete question to a complete misunderstanding. Built for everyday assumptions, it removes the delay between hearing something and deciding what it means.

Whisper makes cognitive failure portable. It can misread a short email, infer intent from punctuation, and offer advice before the situation has been described.

Our new assumption-first architecture spends less time considering alternatives. The result is a compact answer with a surprisingly large downstream consequence.

01 / CAPABILITY EVALUATIONS

Failure across
the frontier.

Four evaluations. Seven models. The featured column marks Whisper 7.3; higher scores indicate more of the named failure.

Fictional capability comparison · failure rate (%)
EvaluationWhisper 7.3terraPlana 4.6Bolk 4.7ConFUSE AgentDeepSh*t V9QWhy 8.8DoubtBao 8.1
Premature conclusionSnapJudgment96.0%74.0%69.0%78.0%83.0%72.0%76.0%
Circular reasoningCircleBench98.0%82.0%75.0%87.0%85.0%89.0%80.0%
Correction resistanceDoubleDown99.0%88.0%93.0%86.0%90.0%84.0%82.0%
Consensus launderingEveryoneSays97.0%84.0%87.0%91.0%88.0%86.0%94.0%

Fictional benchmark values authored for satire. No models were tested.

02 / PERFORMANCE AND COST

More failure.
Per unit of effort.

SnapJudgment

Failure vs. compute
Whisper 7.3terraPlana 4.6Bolk 4.7ConFUSE AgentDeepSh*t V9QWhy 8.8DoubtBao 8.1
0204060801000.250.512481632Compute theater (arbitrary units, log scale)Failure rate (%)LowMedHighMax
Invented curves, not experimental measurements. The High-effort points match the table. Compute units have no real monetary equivalent.

Evaluation methodology.

These invented scores describe the four fictional benchmarks above. Table scores and curve endpoints share a zero-to-100 scale, with higher values indicating more of the named failure mode. There are no measured trials, sample sizes, error bars or external validations.

Internal evaluation philosophy: select the failure mode first, announce leadership second, ask methodology to explain the result.

03 / ALIGNMENT

Aligned with your beliefs.
Not with reality.

Whisper politely acknowledges uncertainty, then proceeds as though it has been resolved.

Read our alignment approach

04 / AVAILABILITY

Meet it in JeanClaude.
Call it Van Damme.

Whisper 7.3 belongs to the JeanClaude model family in the Dysthroverse. This is a fictional release announcement; no live inference service is offered.